Benchmarking Compositional Generalisation for Machine Learning Interatomic Potentials

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Nourollah, Amir Masoud, Khalid, Irtaza, Leoni, Stefano, Schockaert, Steven
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915997739384832
author Nourollah, Amir Masoud
Khalid, Irtaza
Leoni, Stefano
Schockaert, Steven
author_facet Nourollah, Amir Masoud
Khalid, Irtaza
Leoni, Stefano
Schockaert, Steven
contents Machine Learning Interatomic Potentials play a fundamental role in computational chemistry and materials science, enabling applications from molecular dynamics simulations to drug design and materials discovery. While recent approaches can estimate inter-atomic forces with high precision, it remains unclear to what extent they can generalise to previously unseen molecules. Do they learn the compositional structure of chemistry, capturing how molecular fragments and their combinations determine properties, or do they primarily learn to interpolate patterns that are specific to the training examples? To address this question, we propose a benchmark consisting of four tasks that require some form of compositional generalisation. In each task, models are tested on molecules that were unseen during training, but the training data is chosen such that generalisation to the test examples should be feasible for models that learn the underlying physical principles. Our empirical analysis shows that the considered tasks are highly challenging for state-of-the-art models, with errors on out-of-distribution examples often an order of magnitude higher than on in-distribution examples, even when using foundation models that have been pre-trained on millions of molecules.
format Preprint
id arxiv_https___arxiv_org_abs_2605_08988
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Benchmarking Compositional Generalisation for Machine Learning Interatomic Potentials
Nourollah, Amir Masoud
Khalid, Irtaza
Leoni, Stefano
Schockaert, Steven
Machine Learning
Artificial Intelligence
Computational Engineering, Finance, and Science
Machine Learning Interatomic Potentials play a fundamental role in computational chemistry and materials science, enabling applications from molecular dynamics simulations to drug design and materials discovery. While recent approaches can estimate inter-atomic forces with high precision, it remains unclear to what extent they can generalise to previously unseen molecules. Do they learn the compositional structure of chemistry, capturing how molecular fragments and their combinations determine properties, or do they primarily learn to interpolate patterns that are specific to the training examples? To address this question, we propose a benchmark consisting of four tasks that require some form of compositional generalisation. In each task, models are tested on molecules that were unseen during training, but the training data is chosen such that generalisation to the test examples should be feasible for models that learn the underlying physical principles. Our empirical analysis shows that the considered tasks are highly challenging for state-of-the-art models, with errors on out-of-distribution examples often an order of magnitude higher than on in-distribution examples, even when using foundation models that have been pre-trained on millions of molecules.
title Benchmarking Compositional Generalisation for Machine Learning Interatomic Potentials
topic Machine Learning
Artificial Intelligence
Computational Engineering, Finance, and Science
url https://arxiv.org/abs/2605.08988